Who this applies to
This obligation applies to
providers and
deployers of
high-risk AI systems (as defined in
Article 6(1) AI Act) that involve automated decision-making, as well as
AI systems intended to interact with natural persons (per
Article 52(1) AI Act). Public authorities and private entities using AI for employment, credit scoring, or public benefits determinations are explicitly in scope.
What is required
- Transparency for high-risk AI systems: Provide clear, concise, and intelligible information to affected individuals about the AI system’s purpose, intended use, and decision-making logic, as required by Article 13(1) AI Act (co-cited with GDPR Art. 13).
- Right to explanation for automated decisions: Ensure individuals can obtain an explanation of the decision made by a high-risk AI system, including the main elements considered and the logic involved (Article 14(1) AI Act).
- Human oversight mechanisms: Implement procedures to enable human review of automated decisions, as mandated by Article 14(4) AI Act, including the ability to contest or request reconsideration.
- Technical documentation for transparency: Maintain and provide, upon request, documentation detailing the system’s training data, performance metrics, and limitations (Article 11(1)(b) AI Act).
- Real-time disclosure for interactive AI: For AI systems interacting with individuals (e.g., chatbots, deepfakes), disclose the artificial nature of the interaction before or immediately upon engagement (Article 52(1) AI Act).
Key deadlines
The primary deadline for this obligation is
August 2, 2026 (transparency obligations under
Articles 50 and 52 AI Act).
Enforcement patterns
AI Act enforcement begins August 2, 2026. No precedent currently exists. This page will be updated as enforcement cases emerge.
Cross-border considerations
Implementation guidance and early enforcement focus vary by member state, with
Italy (Garante),
Austria (DSB), and
Belgium (APD/GBA) showing higher citation rates for
Article 14 AI Act (co-cited with GDPR Art. 22). No jurisdiction-specific deviations from the AI Act’s text have been documented in enforcement actions to date.